English text is being prepared.
Every rollout, reasoning step and tool call is paid for before you know if it helped. Most don’t. We argue the next leap toward AGI comes not from computing more, but from a cheap System-1 that decides what a costly System-2 should compute at all. https://t.co/6GIRws2YNi
#Papers/Research
#Industry news#Video
Edge AI inference may open a new AI role for telecom operators, and the clearest near-term opportunity lies in metro sites such as central offices rather than at cell towers. EnterpriseWeb CEO David Duggal noted that deploying GPU inference at towers and cell sites faces a power and cooling Catch-22: the higher the power consumption, the greater the heat generated, so some AI-RAN approaches push heavy-load inference back to data centers or aggregation layers. Nvidia has already been advancing the edge deployment of AI-RAN and physical AI in collaborative projects with AT&T, T-Mobile, SoftBank, Telefónica and others.
#English text is being prepared.#English text is being prepared.#Reasoning
The Wikimedia Foundation has weighed in on the copyright lawsuit between OpenAI and the New York Times. General Counsel Stephen LaPorte said that if a ruling applies the fair use doctrine without careful consideration, it could threaten the sustainability of free knowledge and stifle the development of open-source technology. Wikimedia believes AI companies should comply with licensing requirements: using Wikipedia content requires attribution under the Creative Commons license, and any adapted works must carry the same license. CEO Bernadette Meehan said that expanding access to knowledge through AI is an important goal, but it must respect the contributors and communities that make reliable knowledge possible.
#English text is being prepared.#OpenAI
Latent Space host swyx introduces the curation approach for the annual flagship conference AIE CODE, which is single-track, invitation-only, and aimed at about 1,000 agent engineers. This year's three themes are Scaling without Slop, System One models, and Increasing Ambition, among which Jarred Sumner will share Bun's rewrite in Rust and a Bun AOT preview. The article also rounds up that day's AI news, including Mistral's open-weight release and OpenAI's frontier model monitoring.
#Industry news#Agents#English text is being prepared.
Microsoft, Alphabet, and Amazon successively released quarterly earnings reports. Microsoft's Azure revenue grew 48%, with free cash flow positive at $18.6 billion; Alphabet's Google Cloud revenue grew 103%, but free cash flow was negative at $9.065 billion; Amazon's free cash flow over the past 12 months was negative at $42.357 billion. The article cites Goldman Sachs' judgment that the market's focus on AI has shifted from the scale of capital expenditure to revenue backlog, growth rate of change, and the divergence between free cash flow and GAAP operating profit.
#Industry news#Microsoft#Google
On October 30, affected by Google, Amazon, and Intel's earnings missing expectations and weak third-quarter reports from some A-share companies, computing power hardware stocks extended their decline, with Eoptolink and Victory Giant Technology falling over 10%. That day, the Shanghai Composite Index fell 1.44%, the ChiNext Index fell 4.13%, and 4,602 stocks across the market declined. Intel expects personal computer sales to continue slowing, with pressure lasting until 2028.
#Industry news#English text is being prepared.
The first-phase production capacity of Dale New Material's Huizhou company is already running at full capacity, and it is mass-producing the newly launched foldable screen back glass for an international consumer electronics leader, with mass production of new models to gradually increase over the next two years. The company's robot tungsten wire tendon ropes continue to undergo validation with mainstream domestic manufacturers and are expected to generate sizable revenue next year; revenue from products in non-photovoltaic fields already accounts for more than 40%.
#Industry news
Envision Group has been selected for the main list of the "2026 Forbes China Best Employers" for the fourth time, having previously made the list three times in 2022, 2024, and 2025. This year, Forbes China added AI indicator questions to its questionnaire for the first time, covering strategic alignment, AI investment, and AI operations. Envision combines the "Envision Tianji" meteorological large model and the "Envision Tianshu" large model with the EnOS intelligent IoT operating system to build a future-oriented AI power system.
#Industry news#English text is being prepared.
At the 2026 Tencent Global Digital Ecosystem Summit, Maiya Media shared the AI short drama industrialization workflow of Jump Vision in collaboration with Tencent Cloud, reducing the production cycle of a single short drama from about 1 month to 1 to 2 weeks, cutting computing power costs by 40% to 60%, and reducing asset-stage costs by 50% to 70%. Tencent Cloud provides capabilities such as large model invocation, AI image quality enhancement, and seamless erasure; its seamless erasure technology covers more than 80% of leading short drama clients and processes over 50,000 minutes of video per day on average. In the case of "Breakfast Grandma," the film was completed in about 2 days and has been viewed 310 million times.
#Industry news#Video#Agents
The China Iron and Steel Association said on October 30 that the steel industry is shifting from "experience-based steelmaking," which relies on eyesight and feel, to "data-driven steelmaking," which relies on digital technology and AI large models. Baosteel's world's first AI smart blast furnace has officially begun tapping iron, and Hegang's Weisaibo 2.0 full-domain large model, Hunan Steel's Pangu large model, Nangang's Yuanzhi steel large model, and Liugang's Xuantie large model have all been implemented and applied. Ansteel can achieve annual cost reductions of more than 17 million yuan through AI-empowered production management, and Tangsteel has achieved digital application in more than 90% of its process scenarios, with steelmaking efficiency increased by 10.3%.
#Industry news#Embodied AI
Bairong Intelligent released the unified Voice AI brand "Bairong Shengdong" and the voice intelligence foundation VoxOne, and launched three product lines: Baifu silicon-based customer service, Baiying intelligent marketing, and Baiyu voice interaction open platform. VoxOne is based on the self-developed BR-Voice series models, adopts an end-to-end voice interaction approach combined with the Vortex streaming inference engine, with end-to-end response latency within 500 milliseconds, overall response speed more than three times faster than traditional solutions, and inference cost reduced by 30%. Products are priced on a RaaS basis, charged by successful work orders or valid call counts, the number of deployed silicon-based employees, or revenue sharing based on the business scale facilitated.
#Product updates#Speech#Agents
In the third quarter, Amazon's AWS revenue reached $45.984 billion, with year-over-year growth rising to 39.3%, slightly beating the Wall Street consensus of 38% but falling short of the buy-side threshold of over 42%; advertising business growth plunged from 26% to 12%, far below the 18% expectation, dragging the stock down 7% in after-hours trading. CEO Andy Jassy expects AWS to achieve growth of over 40% in both 2027 and 2028, and is expected to reach a scale of $1 trillion within five to ten years, while also planning to sell Trainium chips to customers outside AWS starting at the end of 2027. Jefferies and Goldman Sachs still maintain Buy ratings, with target prices of $330 and $375, respectively.
#Industry news#Agents
Alphabet and Amazon released disappointing quarterly reports in succession this week, and both companies are still maintaining a combined capital expenditure plan of more than $400 billion for the full year. Alphabet's search advertising growth slowed to 12%, the lowest in five quarters, with its stock falling more than 11% in a single day and about $500 billion in market value wiped out; Amazon's free cash flow for the quarter was -$33.5 billion, more than four times the negative figure expected by the market. A Goldman Sachs trader pointed out that the market is shifting from rewarding capital expenditure growth to demanding evidence of returns.
#Industry news#OpenAI#Google
Dongshan Precision's net profit attributable to parent in Q3 2026 was 2.332 billion yuan, up 401.18% year-on-year and up 26.26% quarter-on-quarter, with revenue of 17.582 billion yuan, up 73.81% year-on-year. The growth was mainly driven by the explosive growth of the newly added optical module business following the consolidation of Source Photonics. Net profit attributable to parent for the first three quarters was 5.289 billion yuan, up 332.34% year-on-year. Over the same period, financial expenses surged more than tenfold year-on-year to 824 million yuan. The company's H shares were listed on the Hong Kong Stock Exchange on October 28.
#English text is being prepared.
The Efficient-Large-Model team, composed of researchers from NVIDIA and MIT, has released the model weights for the video enhancement project Open-H3-Regenerate-2K on Hugging Face, with the accompanying inference code now available in the NVlabs/Sana repository. Based on MiniMax H3, the solution regenerates low-resolution video drafts into high-resolution video, supports 2K output at 2560×1440, and can achieve native 4K regeneration at up to 3840×2160. The team kept the H3 Transformer backbone and video VAE parameters unchanged, training only the newly added refinement LoRA; at inference time it is merged with community adapters such as LightX2V FL2VA Turbo and LMS Sharp, regenerating image details through three-step denoising. On a single NVIDIA B200 GPU, the combined time for latent-space upscaling, three-step refinement, and 2K decoding of an approximately 5-second video dropped from 87.26 seconds to 22.45 seconds, about 3.9 times faster than before; on RefinerBench-150, UniPercept averaged 60.0, compared with 53.4 for the official H3-Regenerate-2K API.
#English text is being prepared.#Video#Open-source ecosystem
Axel Scheffler, illustrator of "The Gruffalo," said he has contacted German lawyers and is seriously assessing whether to file a lawsuit over AI reproducing his characters and style, but has not yet disclosed the name of the company he plans to sue. He believes EU regulations are friendlier to creators, and is therefore considering suing in Germany; lawyer Nick Buckland noted that Germany already has a precedent in which a music copyright collective management organization won its case, but generative AI defends itself on the grounds that it trains on works rather than directly copying them, so proving the case remains difficult. OpenAI has already been sued in Germany by Penguin Random House and illustrator Ingo Siegner, and the Munich Regional Court will rule in spring 2027.
#Industry news#Image generation#English text is being prepared.
Yao Maoqing, co-founder of Zhiyuan, said in an interview with Securities Times that general large models have obvious shortcomings in physical robot operations, and future systems will move toward a layered architecture: general large models handle high-level long-horizon planning, world models and VLA output mid-level skills, and the low level is handled by a 100–1000Hz high-frequency motor real-time control layer. He pointed out that a general large model may need 1–2 hours to complete a physical operation and consume a large number of tokens, and that in real-world tests its success rate on tasks involving highly dynamic motion and force/tactile feedback is almost zero. At the 2026 Zhiyuan Innovation Day, Zhiyuan released its new-generation embodied foundation model GO-3 Preview, which adopts the WRAM native architecture and is planned for full open source by the end of 2026; on 100 tasks across 20 skill categories, the model's average zero-shot success rate rose from 13.4% to 56.4%.
#English text is being prepared.#Embodied AI#Model releases
Peng Zhihui, co-founder, president and CTO of Zhiyuan, said at the 2026 Zhiyuan Innovation Day that 10,000 humanoid robots rolling off the production line is only a victory for manufacturing, and that 10,000 robots truly "on the job" is the victory of physical AI. Zhiyuan also released a measurement unit, Worken (effective intelligent work hours), to measure the actual output of robots autonomously and continuously completing valuable work, and hopes to bill by Worken work hours in the future. Peng Zhihui divided the development of physical AI into five stages, believing that Zhiyuan has passed the first stage and is moving toward deployment in vertical-domain scenarios, while the home is the endpoint rather than the starting point.
#Industry news#Embodied AI#English text is being prepared.
AgiBot, together with Tsinghua University and Infinigence AI, has introduced the Collective Experience Learning (CEL) paradigm. Built on the FERT failure-guided experience retrieval algorithm and the FATS fault-tolerant asynchronous training system, it uses a cluster of 96 real robots to enable robots to teach robots, breaking free from a heavy reliance on demonstrations by human experts. In tests on 5 real-world tasks including flower arranging, clothes folding, and phone packaging, the CEL approach achieved an overall success rate up to 16.7 percentage points higher than the average level of human teleoperators, and with the same training duration, the larger the cluster size, the higher the operational reliability.
#Papers/Research#Embodied AI#English text is being prepared.